Ai?

[Zombied00d1]

I know Ai is frowned upon but there should be a specific area for it here. I’ve been messing with it to make celebrities involve in our fetish. Would love to see what others have come up with as i know i can’t be the best only one.

[ZaraSandals]

Hi, I'd be curious to know what others think too. Looks like you've double-posted, could mods please merge the two threads?

As to whether there should be a specific forum - right now given forum activity levels are still low, on balance I'd say increasing the number of forums is not going to help. It's better to concentrate the content in the existing forum structure but encourage posters who use AI to label their content as such.

I don't know if AI is "frowned upon" broadly, within the context of this forum I personally think it's acceptable as long as it's done to a high quality (looks realistic and natural), is clearly labelled as AI, and you're not using it to create and distribute indecent/harmful images of real individuals including celebrities. In one respect it's not so different from posting a fictional shoe cumming story/fantasy, although obviously AI images have much greater visual impact and therefore are more consequential.

My experiments with GenAI are mainly to recreate/improve scenes where my wife wore a certain pair of heels but I hadn't taken any decent photos, or to fantasise about her wearing existing/new heels with different outfits in different situations. Nothing dirty such as nudity or cumming on feet and shoes (so far - but I am inspired to try some time based a previous thread here where someone used prompts with edible products or other cum-like liquids to bypass filters..).

Examples of some of my work in this thread, my wife in grey block-heeled mules:
viewtopic.php?p=1219239#p1219239

Currently I'm having a lot of success with Nano Banana Pro (paid version of Google Gemini), from a quality perspective the AI engines have really come a long way since even a few months ago when you would get deformed limbs, three feet and 6 toes etc. With good quality, multiple source images and suitable specific prompting, I've been able to create very realistic shoe swaps with the shoe design accurately reproduced (including fine sandal straps, unique heel designs etc), changing the pose, posture, and silhouette to account for the height increase with heels, feet dipping and dangling, adding nylons, covering the shoes and feet in sand or water at the beach, and so on. The hardest part so far has been to force the AI to not change the face but there are ways around that too - if blurring the face anyway to protect identities as I do with my wife's pics, it doesn't matter so much.

[ff17]

I would agree on AI content here, because it would definitely pump up the site interest. In general, AI hate is for wrong reasons in many people - rather than hating AI, people should hate those CEOs and decision makers that are greedy and want to make quick buck through AI, or foolishly want to replace people, only to fail miserably.

One can make ANYTHING they want if they run certain AI models locally. The ones that are open source(they can usually be downloaded from CivitAI site, as well as numerous LoRAs).

There is also one site - Perchance - that runs on Flux Schnell AI model and is basically a kind user allowing others to generate content on his own rig. Usually sites have NSFW blocks, not this one.

Aside image generation, video generation can be made too on some other sites (either from text-to-video, or image-to-video), but if made locally, it takes a lot of resources and requires one to have powerful rig for it. Otherwise one can use some other paid services (some of which are not so cheap, but hey...).

But if run locally, one can run image generation through ComfyUI. Time to generate image depends fully on rig, resolution and size of images, AI model, etc...

Some AI models are better for photorealism (like Flux and Z-Image), while some others are better for anime or comics or similar (like SDXL, specifically Illustrious). Any AI model can make anything, but quality will vary and will depend on prompt.

Im in process of transitioning to proper local generation now.

To clarify :

- you have main base AI models, and their checkpoints, which are like slightly modified full versions of AI models.
- you have LoRAs (Low Rank Adaptations) which are like modules that are certain specific content, for example, specific characters, objects or effects, or acts, and then those are "attached" into main AI model, and triggered with keyword.

Those LoRAs are trained content, and one can train anything locally or on CivitAI, using their own images or videos. On CivitAI site, youre going to find a LOT of shoejob, heeljob content, including various heels types and similar. One thing that you wont find - at least not directly - are celebs, which were there before but were removed because of controversies. Still, one can find some on some other places like HuggingFace site (which is like github repo for Ai models and content for it).

So, one can train anything locally. You see that pair of shoes you have images ? You can train AI on them. You see those pair of feet you have been salivating for ages ? You can train AI on it. Even on one image AI can be trained, but recommended number is 20-50, always try to have high quality pics. Because AI will also learn everything related to it - blurry, grainy stuff, artefacts, etc.

One can train also ANY face. Altough thats bit controversial, not without reasons sometimes, depends.

Useful links :

CivitAI site : https://civitai.com/

Hugging Face - one of users making various celebs (you can search for any term in search bar on site) : https://huggingface.co/malcolmrey

Perchance : https://perchance.org/prpg

For Perchance link above, this is just one of numerous generators. All generators use same Flux Schnell AI model, and they only differentiate in UI, which are basically text prompt inputs that are made modularly and can be inserted with selection instead of writing everything in.

Enjoy !